The title says it all, really. Fixed effects in the plm package still estimates coefficients for time invariant variables. Now, if this isn't normal, you might think this is a data entry issue, but I don't think it is a data entry issue. Hausman-Taylor in the plm package correctly distinguishes the time invariants from the time variants using the same data.
So, is this normal? Am I supposed to omit them myself? Do I run the regression including them but excluding their results when I display it to my reader?